All the jmp_ind_x tests contain a movq with an immediate as the location. However immediate values and labels are not valid locations as by the X86lite documentation.

I think there is also a typo on line 110. It says cs_not instead of cc_not

Thanks for sharing your tests!

The Fibonacci test, calculates the 30th Fibonacci number. On our implementation, this gives sometimes a timeout (5s). Maybe lower it to Fibonacci of 20, which works more reliably.

The Fibonacci test times out after 5 seconds. My implementation requires at least a minute to complete the test.
Either my implementation is horrendously slow or yours is quite fast or this test is too ambitious. Could you confirm that you can indeed run this test in the 5 second constraint?

The bitcast_1, bitcast_3 and bitcast_4 tests use the i8 type, which is specified to be not part of LLVMLite - LLVMLite only supports i8* pointers according to the documentation on moodle...

@dbernhard-0x7CD tagging you since you are adding a good quantity of new tests.

I was thinking that if the other assignments are also having tests, it can be good to add an author file maybe? Or license on top of each file?

I would however keep all the file for each assignment in one file only so we can simply copy paste the whole thing to try out our implementation.

Otherwise we can have multiple suites, each of them gathering the test of an author such as:

let nicdard_provided_tests = [ ... ]
let dbernhard_provided_tests = [ ... ]

and then merge all of them (since those are simply list of assertion test).

I believe that the test cc_shr_max tests for the wrong ZF flag.
ZF should be only set (true) iff the result is zero. Int64.max_int in binary is a zero followed by 63 ones.
Shifting right by 63 or more would indeed result in a zero. But the test only shifts right by 3, which clearly is non-zero.
Either the test was meant to shift by 63, test ZF: false or I misunderstood something.

@dbernhard-0x7CD's tests include comparing references, for example, in

https://github.com/nicdard/compiler-design-eth-tests/blob/main/05/dbernhard/equality.oat#L25

I wonder, though, whether supporting this is really necessary for this homework. Although the rules Typ_Eq and Typ_Neq imply that it should be supported, the provided frontend does not implement this (see lines 246 - 252). Also, I did not find any indication anywhere else that we're supposed to add this.

Am I missing something?

Shouldn't this test expect the result to be -2?

It moves 1 to Rax which is 0b00..001. If this is negated bitwise we get 0b11..110 which corresponds to -2.

If i compile this assembler:

.text
.global program

program:
    pushq %rbp
    movq %rsp, %rbp

    movq $1, %rax
    not %rax

    movq %rbp, %rsp
    popq %rbp
    ret

.data  

The Rax register contains -2 as signed int64_t after execution.

Edit: I misunderstood the first argument to Gradedtests.cs_test which is not the expected answer, but the number of instructions to execute. Nevertheless now we have a testcase which tests if the sign flag is unaffected by notq.

I like that the tests for HW3 have a prefix variable for where the files are located.

Could we agree to extend this system with a global prefix?

My goal is to have this repository outside of my personal hw directories but be able to access the tests from there.
If we have one global ml file with a prefix variable, which everybody can configure to a value to their liking, the repo could be cloned anywhere, and from then on, updating would be as easy as a pull. Also, for future homeworks, the only setup one needed to do would be to configure the main and makefile.